Periductal Stromal Collagen Topology of Pancreatic Ductal Adenocarcinoma Differs From That of Normal and Chronic Pancreatitis

Cole R Drifka; Jo Tod; Agnes G Loeffler; Yuming Liu; Gareth J Thomas; Kevin W Eliceiri; Kao W John

Disclosures

Mod Pathol. 2015;28(11):1470-1480. 

In This Article

Materials and Methods

Human Tissue Microarrays

To assess collagen in different pancreatic tissue types, formalin-fixed paraffin-embedded human tissue microarrays were obtained commercially (US Biomax, Rockville, MD, USA). We also obtained additional human pancreatic tissue microarrays constructed from surgical specimens from patients that underwent resection for primary pancreatic ductal adenocarcinoma tumors at University Hospital Southampton. Human tissues were obtained between 2000 and 2010, under approval from the institutional review board (10/H0502/72). All tissues used in this study were derived from tumor excisions with curative intent (not diagnostic biopsies). Cases of hereditary pancreatitis or autoimmune pancreatitis were not included. Hematoxylin and eosin-stained slides were reviewed to confirm the diagnosis, and cases were excluded at this stage if it was unclear whether tumor originated from the pancreatic head or distal common bile duct. One to three cylindrical cores were taken from representative areas of each tumor block. Areas sampled included areas containing tumor islands, areas containing only tumor-associated stroma and surrounding areas with no evidence of tumor involvement as determined by a consultant pathologist.

Multiphoton Microscopy

Multiphoton laser-scanning microscopy was performed using a custom-built workstation at the University of Wisconsin Laboratory for Optical and Computational Instrumentation imaging research facility. All images were acquired using a Nikon Eclipse TE2000U inverted microscope through a Nikon S Fluor 20 × air-immersion objective (numerical aperture=0.75) (Nikon, Chiyoda, Tokyo). A mode-locked MIRA 900 Titanium:Sapphire laser (Coherent, Santa Clara, CA, USA) was tuned to an excitation wavelength of 890 nm to deliver ~10 mW of power at the sample. A Semrock 445±20 nm narrow-band pass filter was used to isolate the backscattered second harmonic generation signal. Tissue microarrays were imaged in entirety using an acquisition grid defined in WiscScan, a laser-scanning software package developed at the Laboratory for Optical and Computational Instrumentation (https://loci.wisc.edu/software/wiscscan). Individual images of 512 × 512 pixels were acquired within the constraints of the grid with a 10% overlap between images. Following acquisition, a FIJI plugin[22] was used to stitch the images together based on OME-XML metadata.[23,24] Two-photon excited fluorescence images were also acquired without filtering in selected regions of interest.

Pathology Review

Hematoxylin and eosin-stained whole-tissue microarray cores were designated by pathology as either representing normal pancreas, chronic pancreatitis only (derived from pancreatic ductal adenocarcinoma cases but with no malignant elements present), or pancreatic ductal adenocarcinoma. Carcinoma tissues were assigned a histological grade according to the three-tier grading scheme by two pathologists. In total, tissues from 241 patients were considered. Select tissues (from 117 patients) were subjected to additional review by a pathologist (AL) to assess collagen changes in relation to relevant epithelial–stroma interfaces. Tissue microarrays were digitalized using an Aperio CS2 scanner system (Leica Biosystems, Buffalo Grove, IL, USA). Using ImageScope viewing software, normal ducts, ducts in chronic pancreatitis, and pancreatic ductal adenocarcinoma ducts were identified and annotated (1–3 annotations per core). Each annotation was the same size (400 × 400 μm), and representative regions were chosen where roughly a 1:1 epithelial–stroma proportion existed. The pathologists were blinded to the second harmonic generation data, thus eliminating the possibility that collagen visualization influenced how the tissues were reviewed and annotated.

Computational Collagen Fiber Segmentation and Quantification

Collagen fiber quantification was done using CT-FIRE, an open-source software package developed to automatically segment and quantify individual collagen fibers from second harmonic generation images (https://loci.wisc.edu/software/ctfire).[25] CT-FIRE was designed to compute the overall alignment of collagen fibers as well as individual length, straightness, and width. These fiber features were chosen because they appear to be altered in many cancer types compared with normal tissue counterparts.[19] Fiber length and width are calculated as pixel values. Alignment represents the overall directionality of fibers within the image on a scale from 0–1, where 1 indicates all fibers are orientated at the same angle. Straightness is calculated by dividing the distance between each fiber end point by the distance along the path of the fiber and is also on a scale from 0–1, where 1 indicates a perfectly straight fiber. For whole-core CT-FIRE analysis, individual stitched cores were first cropped using a 1.625 by 1.625 mm square region of interest tool in FIJI. For annotation CT-FIRE analysis, the regions drawn on the Aperio scans were transferred to the stitched cores and used to crop the second harmonic generation data prior to analysis. All images were eight-bit and thresholded 10–255 to eliminate background noise before running CT-FIRE using default parameters.

Alpha-smooth Muscle Actin Immunohistochemistry

Automated immunostaining (Ventana XT, Ventana, Tucson, AZ, USA) was performed on representative pancreatic ductal adenocarcinoma tissue microarray cores from 39 patients for alpha-smooth muscle actin (1A4, Dako) in an accredited clinical cellular pathology department per the manufacturer's instructions as previously described.[26] The slide was digitalized using an Aperio CS2 scanner system and the Positive Pixel Count algorithm (v9) was used to quantify alpha-smooth muscle actin positivity in the individual cores.

Statistical Analysis

A linear mixed effects model with subject-specific random effects was developed in R statistical software (v3.1.1, www.r-project.org) and used to examine the association between tissue assignment and collagen fiber parameters. The results were summarized in terms of adjusted means and standard errors. Receiver operating characteristic curves were generated using MedCalc (v14.10.2, www.medcalc.org). All statistical tests were two-sided, and a confidence level of 95% was used to establish statistical significance.

processing....